Brown University professor Stefanie Tellex and colleagues have developed an approach that enables robots to quickly determine the sequence of actions that will work in a particular environment. The approach is based on an algorithm that enables robots to prune away certain possible paths of action by understanding the direction in which a particular task points.
The team used the popular computer game "Minecraft" to test the approach, and they report the algorithm controlled a character, learned certain behaviors, and worked through a much smaller set of potential scenarios. The researchers controlled an avatar tasked with putting a virtual gold block into a virtual furnace, while avoiding a virtual pool of lava. After performing the task in a limited setting, the algorithm controlling the avatar learned that certain behaviors, such as placing gold blocks on the ground, could be excluded when trying to achieve the goal.
The researchers also tested it on an actual robot, and they say their approach is more efficient and even more human because it requires a deeper understanding of a task and its context. They note the strategy could be important as robots take on more complex, open-ended tasks in less structured settings.
From Technology Review
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA